v0.5 · live on npm as @meterbility/cli

The debugger for AI agents.

Capture every run. Inspect every decision. Pause and inject live. Fork from any step. Diff the trajectories. Meterbility turns agent runs into a queryable, replayable, forkable corpus.

$npm install -g @meterbility/cli

Works with · Claude Code, Codex CLI, Cursor, Anthropic + OpenAI proxies, TypeScript & Python SDKs

meter — zsh
meter ingest claude-code --limit 5 ingested 5 sessions · 342 steps · 1.2M tokens meter list run_9f3c claude-code fix-auth-bug $1.84 96 steps run_a210 claude-code add-billing-page $0.62 41 steps meter fork run_9f3c --step 44 --continue live forked → run_9f3c.fork1 (deterministic prefix, live suffix) meter diff run_9f3c run_9f3c.fork1 step 44: tool_call diverged — Edit(auth.ts) vs Bash(git revert) meter web inspector live at http://127.0.0.1:4317
01

Browser DevTools, mapped onto agents.

You already know how to debug a web page. Meterbility gives your agents the same panels — all shipping today.

01ElementsContext ViewerThe fully resolved context for any step — system prompt, history, tool schemas — at /contexts/:id.
02SourcesStep Inspectormeter inspect walks every decision; meter fork branches a new trajectory from any step.
03NetworkI/O InspectorDecision blobs, tool results, and files changed — including Bash side-effects other tools can't see.
04PerformanceCost TimelinePer-step and per-run cost, token, and latency timelines. Know exactly where the money went.
05ConsoleLive ProbePause a running agent, inject a message, resume — meter probe or the web panel.
06gitTrajectory Diffmeter diff compares two runs step-by-step and pinpoints exactly where behavior diverged.
02

Full-fidelity capture. Local-first storage.

Everything runs on your machine: SQLite plus content-addressed blobs. Add Postgres when you need multi-machine sync.

F01

Capture every surface

Claude Code JSONL hooks, Codex CLI & Desktop, Cursor composer, and Anthropic/OpenAI proxies via meter proxy.

F02

Bash side-effect capture

What tool-call inspection can't see — sed, mv, build scripts — lands as file-change rows via hooks or the FileSentinel.

F03

Fork & replay

Deterministic prefix replay with a live suffix. Continue with --continue simulate|live for multi-step what-ifs.

F04

Regression suite

Promote a good run to a canonical, add assertions, and catch behavioral drift with meter test.

F05

SDKs for custom agents

@meterbility/agent for TypeScript and meterbility-agent for Python (stdlib-only). One-line Anthropic tracing.

F06

Sensitive-path redaction

Edits to .env, keys, and credential stores record the fact of the change — path, op, size — never the contents.

03

Open core. Start free, scale when your team does.

The full single-operator product is MIT-licensed, forever. Paid tiers add multi-tenant fleet orchestration and the hosted cloud.

Open Source

MIT license

$0

free forever, run it anywhere

  • All capture surfaces (Claude Code, Codex, Cursor, proxies)
  • Terminal + web inspector, Live Probe
  • Fork, replay, diff, regression suite
  • TypeScript & Python SDKs
  • Local SQLite + optional Postgres sync
  • 1,290+ tests, zero copyleft deps
Get started on GitHub

Team

Elastic License 2.0 (ee/)

Early access

join the design-partner program

  • Everything in Open Source
  • Multi-tenant fleet orchestration
  • SSO & role-based access control
  • Audit logs
  • Long-retention storage modules
  • Direct line to the maintainers
Sign up for Team

Enterprise

Commercial · hosted cloud

Custom

tailored to your fleet

  • Everything in Team
  • Hosted cloud — zero-ops backend
  • Compliance & procurement support
  • SLAs & priority support
  • Custom capture adapters
  • Shape the roadmap with us
Sign up for Enterprise

Built in the open.

MIT-licensed core, permissive dependencies only, and a fresh-laptop test script that proves the install path on every CI run.

1,290+tests, two runtimes
11@meterbility/* packages
0copyleft deps
v0.5shipping now
Star on GitHub